GPU's in the Datacentre you say 🤔

gpu Jan 27, 2022

Graphics processing units or GPU's have been around for quite some time but not necessarily in the Datacentre. They can be used to dramatically enhance the performance with a range of  workloads from Artificial Intelligence (AI), Machine Learning (ML), to graphics heavy processing, CAD, rendering, and of course gaming.

So whats the big deal with GPU's anyway 🤷🏼‍♂️?? Well, were a CPU processes pieces of data individually or sequentially, GPU's can process large arrays of data similtaneously or in parallel. This makes them highly efficient at performing certain tasks, especially if it involves crunching lots of data quickly. Check out this cool video that compares them 🤯 - Mythbusters demo of GPU vs CPU.

With the oscillating cycle of system bottlenect jumping between CPU and Memory, there's also the added benefit of reducing your ever busy CPUs workload by offloading what you can to GPUs - winning!

Here at Softsource vBridge I'm excited to announce a trial we'll be running on some impressive new hardware, the Dell PowerEdge R750xa - a purpose built host designed to run GPU workloads.

Loaded up with a pair of Nvidia A40 "Ampere" GPU's which are the latest generation of Datacentre GPUs from Nvidia. Each with 48GB GDDR6 memory and  jam packed with 10,752 CUDA cores, 336 Tensor cores and capable of 696 gigabytes per second (GB/s) of memory bandwidth! These are interconnected via NVLINK to with allows inter GPU communication at high speeds for applications that can take advantage of it. Workloads like high end design, batch rendering, virtual workstations, deep learning training as well as inference workloads would be its bread and butter. The A40 GPU supports the four editions of NVIDIA virtual GPU software: NVIDIA Virtual Workstation (vWS), NVIDIA Virtual Applications (vApps), NVIDIA Virtual PC (vPC), and NVIDIA Virtual Compute Server (vCS).

On the software front it'll be paired with the latest flavour of VMware vSphere 7 (thats update 2 at the time of writing with the unfortunate state of update 3  😢). Additionally we'll be looking to utilise the VMware Blast Extreme display protocol as part of the solution.

So what are we hoping to test/achieve with this host? Well, as always the answer is "it depends"........ initially we'll be working to validate the ability to deliver (and scale) professional graphics type workloads in a multi-tenant environment using the NVIDIA virtual workstation (vWS) solution. Performance will be one of the key measures. The specific workloads will include batch rendering, modeling, Computer Aided Design (CAD), and Computer Aided Engineering (CAE) work.

Another key benefits of hosting GPUs in the Datacentre is of course mobile users ....... Users are no longer tied to an expensive workstation in the office and can work from anywhere as long as they have a good internet connection and an access device! That's probably a handy feature to have in times like these 🚦(did someone say lockdown🙊??).

Anyhow, as mentioned earlier there are a range of potential use cases for GPU's in the Datacentre and this really is only the start. We have racked and stacked the hardware in the DC and are ready to get stuck in with this first use case.

Exciting times and something I'm looking forward to working with over the coming months - hopefully something that will turn into a product offering (or two) we can share with you all. Of course, if you have a candidate workload the might be able to make use of our GPUs sing out - we'd love to help. 👋🏼